Crate vikos

A machine learning library for supervised regression trainings

This library wants to enable its user to write training algorithms independent of the model trained or the cost function tried to minimize. Consequently its two main traits are currently Model and Cost. The two submodules model and cost provide ready to use implementations of said traits.

Modules

cost	Implementations of Cost trait
linear_algebra	Defines linear algebra traits used for some model parameters
model	Implementations of Model trait

Traits

Cost	Cost functions those value is supposed be minimized by the training algorithm
Model	A Model is defines how to predict a target from an input

Functions

gradient_descent_step	An SGD training step with a velocity term
inert_gradient_descent_step	Changes all coefficents of model based on their derivation of the cost function at features
inert_stochastic_gradient_descent	SGD tranining with constant learning rate and velocity
stochastic_gradient_descent	Applies a plain SGD training step to model once for every event in history using a constant learning rate